Google self-driving car hit again, so heads up, Austin drivers

SAN FRANCISCO — If Google’s self-driving car test program is proving anything, it’s the obvious: Humans can’t drive as efficiently as computers. And those computers are only getting smarter.

That would  appear to be the explanation for yet another rear-end collision suffered by a Google autonomous car last month, as reported by the company this week in its ongoing monthly updates on its 6-year-old program.

Other news in the blog post included the announcement that a few of Google’s pod-like prototypes will soon join the company’s gizmo-strapped Lexus SUVs for extensive testing on the streets of Austin. The bulk of Google’s testing has been near its headquarters in Mountain View, Calif.

Among the challenges facing Google’s new Texas-based fleet are dealing with an abundance of deer and processing information from horizontally positioned traffic lights.

Google has tested its autonomous cars for more than a million miles in an effort to pack its database with as many real-world driving situations as possible.

Hitting the streets has meant getting hit. Google’s self-driving cars have been in 16 accidents, 12 of them rear-end collisions and all of them the fault of human motorists, according to data the company provided to the California Department of Motor Vehicles.

The latest took place Aug. 20. A self-driving Lexus was approaching a Mountain View crosswalk without a stoplight when it detected a pedestrian stepping into the road. The autonomous vehicle immediately began braking, but the onboard safety driver decided to intervene and brake more firmly.

As the car stopped, it was hit from behind by a motorist  busy changing lanes at the same time. Typically, such accidents are deemed the fault of the person doing the colliding, as he or she should have allowed for more of a gap between the vehicles.

Program leader Chris Urmson noted that after analyzing the accident data,  engineers determined that although Google’s human driver followed protocol by taking control in an emergency situation, had the car been left to brake on its own, it would have done so less violently and edged  closer to the sidewalk.

“In other words, our software might have created some extra margin in a situation where fractions of inches and seconds mattered,” Urmson wrote. “(Software) could do the math on many complicated factors all at once … and make an extremely nuanced braking calculation. … We’re putting this down as an officially interesting moment in self-driving car history.”

The August report mentions that Google engineers are working on how to tell self-driving cars where to go once a passenger is seated inside the two-person prototype. “People are already carrying their phones, so it would make sense to take advantage of that,” Urmson wrote.

As for when Google’s fleet might hit the roads for real, the post reiterates (nervous Dad) Urmson’s determination to make sure his son doesn’t need to get a driver’s license. He’s 11 now, so in about four years, prepare for the pod invasion.